Why I recently wrote an article in note —— the story of creating a "public candidate" column in the daily report
Although we are helping companies introduce AI, there was a time when AI was adding more work to me.
It's not that using AI makes it easier.When I started working on the premise of using AI, I was busier than ever.I can still write the article.Write down why.
There was a time when I was rather busy when I started using AI
I'll tell you first.This is not to say that using AI will make your work easier.
There was a time when I used to post articles in my note every day.Since I started using it to write AI, the number of articles has increased.However, the feeling of being busy has rather risen.Now that I can write articles with AI, the total amount I can write has increased.With the increase in the total amount of writing, there was more work to be done.
According to a survey conducted by Persol Research Institute (February 2026), “61.2% of the time saved has been reinvested in work.”Even if you make time with AI, that time will be filled with work again.In addition, there is paradoxical data that the more frequent users are, the more overtime they work.
This is consistent with my experience.It's not that I started using AI and had more work, but that I was busier than ever before when I started working on the premise of using AI.By increasing the upper limit of what can be done, the upper limit of the amount to be done has also been raised.
However, even in that busyness, the state of "being able to write articles" continues.Because the work of writing the article itself is not a burden.While AI has increased its busyness, there is a mechanism that has reduced only the cost of writing articles.That is the "public candidate" column of the daily report.
Previously, there was a state of "no information" on a regular basis
It may sound strange to feel that there is no story in the article even though you are doing AI-related work every day.But in fact, it will be like that.
I can't look back because I don't have a record of what I did.I can't see the pattern because I can't look back.After finishing with "You did a lot today," it's unclear what you did a few days later.Only the impression remains, and the specific content disappears.
When you want to write an article, you have to think about what to write.Even if you consult with AI, "What theme are you good at?""Tell me what you've done recently."Without context, of course.It was uncomfortable to ask this question every time.
I need to organize myself to explain what I've done recently.Organizing involves digging up memories.The details don't come out because the memory is vague.If the details do not come out, the instruction to the AI will be "Write an article on the use of AI".With such instructions, only generalities like those that have been read elsewhere emerge.
Many people cite "I don't have time" as the reason I can't write, but it was different for me.The cost of "deciding what to write" was high.Once I started writing, I could do it in an hour or two, but it took me days to decide what to write.The judgment cost far exceeded the start-up cost.
I made a "public suggestions" column in the daily report
After the Claude Code session (work) is completed, a mechanism is in operation to summarize the work of the day as a daily report.The session log is automatically accumulated and organized as a daily report.
Added "Public Suggestions" column here.
# # Public Suggestions| Title Proposal | Recommendation | To | |-----------|--------|--------| | 00Design | | Zenn ⭐⭐ | | Operating△△ with AI | ⭐ | note | ```
The AI reads the contents of the daily report and writes down as candidates what it determines to be "this will be an article".Also set the recommendation and the post destination (Zenn, note, Qiita).
With this alone, the question of "what to write" almost disappeared.
Why does this work?
It is a technical specification that AIs do not take over memories across sessions.Talking about yesterday's work in today's session, I said, “What was it?”It starts with.The context is reset each time.
Most AI services are designed to start from scratch every time a conversation starts.There is no such thing as "what I said last time" or "what I did last week".What I politely explained yesterday was also blank for the next day's session.This is not a low-performance or malfunctioning design.
Then we can create a separate mechanism for passing information across sessions.Daily reports play that role.
If you look at the public candidate column of the daily report, you can start article production with the instruction "Write the article of the candidate last week".Since the context to be given to the AI is clogged in the daily report, it is not necessary to explain "what kind of work was it" every time.Since the details of the work, the background, and the reason for the judgment are recorded in the daily report, the instruction to the AI becomes a pinpoint.If there is a candidate title "What I noticed in the design of" and the contents of the daily report, the AI can consider the structure of the article.
If the AI doesn't remember, put out a mechanism that works instead of memorizing.That's all there is to it.
It became AI's job to decide whether "it can be made into an article" or "it can't be made into an article"
The actual utility of the public candidate column is that it was able to delegate "whether it can be used as an article or not" to AI.
I used to make this decision myself."Is this general?" "Is it useful to others?" "Is there enough content to write about?" - I asked this question every time, and I answered it every time.This was more exhausting than I expected.
Since the details of the work of the day are left in the daily report, the AI can determine "Is this a reproducible knowledge?" and "Will it be a reference for other engineers who encountered the same scene?"The discovery that it was an OS specification, not a problem with my code, becomes an article.I was stuck for 30 minutes due to a misconfiguration, but I was able to quickly identify the cause, which is not an article.The AI will draw this line for us.
Once the judgment criteria are written in the daily report template, the criteria will work in the daily daily report generation.Instead of writing in the system prompt, write in the path (template) that you must take every day.This is the difference between "injecting" and "letting it exist as a structure".
“What I did” and “Why I did it” are separate records
I began to write daily reports, and the way I observed the work changed.
"What you did" is close to an effort record."Fixed configuration file" "Added test" "Updated document".Even if you just write this, it will not be the material of the article.
The material of the article is "why it happened" and "how to deal with the same scene".
In one day's work, the scheduler of the external tool was significantly slower than the expected interval.When investigating the cause, the function of the state record was called before the task was executed, and the record continued to be updated even if the process was skipped.I corrected it with the pattern "Write it down, skip it, and return it".After correction, it fell within approximately ± 5% of the target value.
There is a record in the daily report that it is not "I fixed the scheduler" but "When the timing of skip processing and status recording shifts, it becomes a cumulative bug, and it is solved by the tentative writing + revert pattern".The latter will be the article.
The reason why we began to write the daily report as an "observation note" instead of a "record" is because we created an open candidate column.In order to submit a public candidate, it is necessary to write "why it happened" in the daily report.Purpose lifted the record.
The relationship between the "check all" trap that is easy to fall into with the use of AI and the daily report
There is a pattern of repeated viewing in the field of AI introduction support.
The operation of "confirming all" the output of AI will become established.Read the full text written by AI.Check all.The person in charge will feel that "I am using AI but my work has increased".I won't be using it anytime soon.
According to a survey by Persol Research Institute (February 2026), only about 25.4% of users actually reduced their working hours with AI.In the JUAS Corporate IT Trends Survey 2025, 4.0% of companies significantly exceeded expectations.The data shows that the majority is in a state of "using but not seeing the effect".
If the operation "I read everything just in case because it is written by AI" becomes established, the experience of the person in charge will be "I do not reduce my work even though I am using AI".I can't trust the output of the AI, so I read them all.If you want to read it all, it's faster to write it yourself from the beginning.As soon as that feeling emerges, the use of AI in the organization stops.
The root of the trap is that there is no definition of where the AI should stop.Since there is no definition, the person in charge falls to the safe side.If you fall to the safe side, all will be confirmed.Checking everything will continue to consume the time that the AI should have increased.
The open candidate column of the daily report is your own answer to this question.
"Decision to choose a candidate for publication" is AI."The final decision on whether to write an article" is human.Generate Article Draft is AI."Confirmation and correction before publishing" is human.I've decided where each of them will be.There is a separation between when the AI should stop and when it can proceed.
Determining your location reduces the cost of confirmation.Instead of checking the full text, just look at what you need to decide.When the items to be checked are narrowed down, a sense of confidence in the AI's output is also generated.The sense of security that "you don't have to look here" allows you to continue to use it.
"I can't write" disappears when the system stabilizes
Looking back on the previous state, there were multiple reasons for "unable to write".
I don't know what to write.Even if you start writing, the material is thin.Even if you try to ask the AI, you have to explain from scratch every time.There is a lot of confirmation work and it continues to be revised to the end.
We were looking for a separate solution for each, but by creating an open candidate column in the daily report, we solved it collectively.
The daily newspaper will tell you what to write.As the number of daily candidates accumulates, it becomes a state of "there are five candidates this week".Instead of "there is no story", it becomes a question of "which one should I write first".
The context of the beginning of writing is clogged with daily reports.You can start with the instruction "Make this work on October day an article".Since the background of the work details, discovery, and judgment of the day is recorded in the daily report, the cost of explaining it to AI is greatly reduced.Articles are born not from general theories but from my own specific observations.
The amount of confirmation work has also changed."Why it happened" is recorded in the daily report, so you only need to confirm the facts of the article written by AI with your own records.Rather than checking from scratch, it will be checked against the daily report.The lower the cost of verification, the more unstoppable the pre-launch review will be.
Many people who do not write feel that they do not have time.However, the reality is often either "it is expensive to decide what to write" or "it is difficult to gather the necessary context to start writing".The daily report solves both.
Writing an article has changed from "starting from scratch" to "forming what is in the daily report".The difference was bigger than I thought I'd try.
Creating a mechanism is not the same as being careful
It does not work even if you say to the AI, "Keep a note of the material of the article from the next time".Because I don't remember them across sessions.
The idea of "creating a path that can't fly" rather than "getting people to be careful" repeats itself in the design of AI utilization.If there is a public candidate column in the daily report template, the candidates will naturally appear in the daily daily report generation.Even if the AI does not remember, the mechanism replaces the memory.Since the template is a path that must be taken every day, it is possible to “make it exist as a structure” rather than “inject” the instruction.
The same idea applies to how Cursor manages project-specific rules in Git, and Devin's design of sharing coding knowledge across projects.Instead of being remembered by the AI every time, create a structure that goes through the same path every time.Instead of letting them remember, put a mechanism outside to replace memory.
How it works
Considering the reason why the open candidate column of the daily report is working, several conditions can be seen.
First of all, I added it with the motivation to write a daily report already in place.It's not about creating entirely new habits, it's about adding to what we're doing.It continues because the cost is almost zero.
Next, the AI should be responsible for the output of the public candidate column.You don't have to decide for yourself if this is going to be an article.The AI reads the daily tasks and selects the candidates.All I had to do was write a daily report, and I left the selection of candidates to the AI.
And even if there are candidates, there will be no penalties.Even if there are 10 candidates, just write 3.There is no sense of obligation to accumulate.This largely determines whether or not this continues.As soon as it becomes an obligation, the mechanism becomes heavier.
A larger mechanism is followed by a smaller one.It works because it continues.The habit of "writing articles for 30 minutes every day" does not continue, but the mechanism of "filling out public candidate columns while writing daily reports" continues.
That's why I've been writing articles in note lately.When you write a daily report, the feeling of the cost of "writing" changes.
Reference: - [Persol Research Institute "Research on how generative AI works"] (https://rc.persol-group.co.jp/thinktank/data/generative-ai/) - [Juas "Enterprise IT Trends Survey 2025"] (https://juas.or.jp/cms/media/2025/02/it25_2.pdf) - [estie "Human-in-the-loop"] (https://www.estie.jp/blog/entry/2025/05/29/094554)
---
*Machine-translated (MyMemory API) from a Japanese original at [nomuraya-hub.pages.dev](https://nomuraya-hub.pages.dev/). Pre-review draft. I am the same author writing under different pen names — "nomuraya / shimajima / 中翔" — depending on the medium.*